Cross-Layer Secure Transmission in Underwater Acoustic Sensor Networks via Dynamic Psychological Game Theory

Abstract

Underwater acoustic sensor networks (UASNs) are vulnerable to cross-layer collaborative attacks, in which network-layer internal threats and physical-layer jamming jointly endanger data delivery, while existing security schemes relying on isolated single-layer defenses are easily defeated by such intelligent adversaries. This paper proposes DPG-SS, a cross-layer secure transmission scheme based on dynamic psychological game theory. In the network layer, a Q-learning routing algorithm fuses energy and trust evidence to dynamically avoid malicious nodes. In the physical layer, a dynamic psychological game solved by counterfactual regret minimization (CFR) predicts the attacker’s jamming intention from historical interactions and adaptively switches between multi-hop and AUV-assisted transmission. Simulation results show that DPG-SS reduces the average attack probability by 28.6%, 35.1%, and 37.0% compared with the QMCR and PCAQR baselines under malicious ratios from 0.1 to 0.3. Ablation studies confirm the contributions of both components: without the psychological game, the baseline attack probability rises to about 0.34, while DPG-SS achieves 0.12; without trust-aware routing, standard zero-sum and evolutionary game baselines achieve an attack probability of 0.068, while DPG-SS achieves 0.022. Sensitivity analyses of the delay threshold and the psychological weights verify the robustness of the scheme under different parameter settings, and complexity analysis confirms its feasibility for resource-constrained underwater environments.